Overview

Stillvault is human-in-the-loop secret management for AI agents and the systems they operate. Secrets sit at rest — sealed — and nothing is released until a named human approves it. Approval is the event, not the exception.

The vendor-blind guarantee

Your secrets are encrypted before they reach us and can only be opened by the approvers you enrol. Stillvault stores ciphertext and public keys; we hold nothing that can read a secret. A subpoena to us, a breach of our database, or a rogue operator on our side yields no plaintext — there is nothing on our side to decrypt with.

This is the property the whole product is built around: even we can’t read your secrets.

How it fits together

• Secrets live under hierarchical /path/based/ ids — db/prod/password, stripe/live/key — so policy can be written once across many of them.
• Approvers are people. They countersign releases from the web console or their phone; the act of approving is the only thing that lets a secret out.
• Consumers — a server, a CI job, or an AI agent — request a secret by path. The request waits until the quorum approves, then the plaintext is delivered to the requester and to no one else.
• Policies decide, per path, who may ask and whose approval is required.

Where to go next

• How it works — the end-to-end release flow.
• Quickstart — set up an org and release your first secret.
• Storing secrets and Approving releases — the two halves of the loop.
• Approver groups and Access policies — governing who approves and who may ask.